Cancer is one of the major threats to public health in the United States, and its etiologic research is urgently needed. Among all of environmental factors, diet has been identified to be one of the major contributors, accounting up to 40 to 60% of the cancer incidence. More importantly, diet is modifiable. Therefore, responsible dietary factors to cancer can be used for cancer prevention and control, and thus, have a significant public health implication. Most of the epidemiologic studies on diet and cancer often adopt matched and unmatched case-control designs, and use logistic regressions for data analysis. Unfortunately, logistic regressions often fail to yield interpretable results because dietary nutrients are often highly correlated, are always contaminated by measurement errors, and often fail to satisfy the logistic function form. New statistical methods are needed to improve the quantification of the association between diet and cancer. The broad objective of this proposal is to establish statistical methods that are particularly useful to but not restricted to studies on diet and cancer. More specifically, regressing an exposure variable on case/control status after adjusting for the other covariates, one establishes a class of new methods that can be used for case-control studies in general. These methods will be particularly useful for studies on diet and cancer, since they require a weaker assumption than logistic regressions, take measurement errors into account and yield stable estimates in the presence of highly correlated covariates. This proposal has six specific aims. The first aim is to establish methods for matched and unmatched case-control studies, which yield the mean difference of exposure between cases and controls with and without adjustment for covariates. The second to fourth aims are to develop methods for matched and unmatched case-control studies in the presence of measurement errors with no validation data, with repeated measurements and with validation data, respectively. The fifth aim is to investigate the similarities and differences between these new methods and other established methods including logistic regression. The six aim is to develop computer programs for these new methods and to make them available to epidemiologists. Development of all these new methods is on the basis of the method of the estimating equations which have been developed in the last five years. To explore the finite sample properties of these new methods as well as to compare them with other proposed methods, we plan to conduct Monte Carlo stimulations. Finally, we will also conduct preliminary and final analyses for the ongoing matched case-control studies on diet and lung cancer by our group as well as analyses for the other case-control studies conducted in our Program Project. Programs for these new methods written in GAUSS will be developed and will be made available to epidemiologists on request.